Corresponding author: A. Gallego, Department of Geology and Geophysics, University of Hawaii, 1680 East-West Rd., Post 813, Honolulu, HI 96822, USA. (email@example.com)
 We located continuous seismic tremor with coherent amplitude wave trains in the Chile ridge subduction region (~46.5°S) in two clusters north and south of the Chonos Archipelago, between the Chile trench and the North Patagonian fore arc. Tremor persisted from December 2004 to February 2007 (the entire period of the Chile Ridge Subduction Project temporary seismic deployment), and lasted >17 h on six occasions. Tremor in the more active southern cluster reached a maximum duration of 48 h, and we observed no more than 3 continuous days without tremor activity. The cluster locations coincide with the surface projections of subducted transform faults formed at the Chile ridge. We also detected simultaneous, colocated low-frequency microearthquakes with well-defined impulsive waves within the tremor signals distributed from the surface to 40 km depth, suggesting tremors and earthquakes are part of the same process. The periodicity of tremor duration is strongly correlated with semidiurnal, diurnal, and long-period tides, M2, N2, K1, O1, P1, and Mm (12.421 h, 12.000 h, 23.934 h, 25.819 h, 24.066 h, and 27.555 days, respectively). We found a significant correlation between tremor occurrence and Earth tides when tidal stress is calculated for the slip plane of a right-lateral strike-slip fault with strike N95°E, which is near parallel to subducted transform faults (N78°E) of the Chile ridge, indicating that the very small stresses resulting from the combination of ocean loading and solid Earth tides (~1 kPa) are sufficient to facilitate or suppress tremor production; tremors occur when shear stresses are maximum and wane or are low when shear stresses are minimum.
 Seismic tremor signals are typically found in volcanic areas and are associated with the underground movement of magma, fluids, or gas [e.g., Chouet, 1996]. However, a type of seismic tremor clearly unrelated to volcanic arc activity—termed nonvolcanic tremor (NVT) —has been spatially and temporally correlated with geodetically detected episodes of slow slip in subduction zone fore arcs [Dragert et al., 2001; Obara, 2002; Rogers and Dragert, 2003; Obara et al., 2004; Obara and Hirose, 2006; Ito et al., 2007]. These events were interpreted as episodic unlocking and slow (i.e., weeks) elastic slip of the plate interface, with concomitant fore-arc surface motion, and the tremors were considered to be either a direct manifestation of fluid release due to dehydration of the slab causing cascading ruptures on small fractures [Obara, 2002] or shear slip on the plate interface mediated by high pore fluid pressures [Rogers and Dragert, 2003; Shelly et al., 2006, 2007a]. Correlation between tremor and soli-lunar tidal variation have been observed in the Japan and Cascadia subduction zones [Shelly et al., 2007b; Nakata et al., 2008; Rubinstein et al., 2008; Lambert et al., 2009] and along segments of the San Andreas fault [Thomas et al., 2009], suggesting that the stresses necessary to trigger tremors are much smaller than stresses involved in earthquake production and that the faults involved are very close to failure. Similarly NVTs were also observed to have been triggered by passage of seismic waves from distant earthquakes [Rubinstein and Vidale, 2007; Gomberg et al., 2008; Peng and Vidale, 2008; Peng and Chao, 2008].
 The geometry of the Chile triple junction (CTJ) and Chile ridge (Figure 1) results in the formation of slab windows [Breitsprecher and Thorkelson, 2009; Russo et al., 2010b]: as the ridge segments subduct below the South American plate, spreading and seismic slip on subducted transforms continues [Murdie et al., 1993] without generating a new crust, as mandated by low-pressure and high-temperature upper mantle conditions, opening gaps between the subducted Nazca (NZ) and Antarctic (AN) slabs. The CTJ region appears to have formed the southern limit of coseismic rupture of the great 1960 Mw = 9.5 Chile earthquake. One aftershock of this event, the 6 June 1960 Ms 6.9 earthquake, occurred in the CTJ region and was a highly anomalous slow earthquake, which produced a 1.5–2 h wave train of anomalously large-amplitude 10–25 s period Rayleigh waves, lacking associated body wave phases, radiating from a hypocenter along the subducted Taitao transform fault [Kanamori and Stewart, 1979].
 The occurrence of NVT in Patagonian Chile has been previously reported; tremor was clearly located in the fore arc, almost exclusively north of the CTJ in the Chonos Archipelago [Gallego, 2010; Ide, 2012]. Here we present results from 2 years of seismic observations of NVT in Patagonian Chile, where the actively spreading Chile ridge subducts beneath South America (Figure 1) [Russo et al., 2010a, 2010b]. We show that NVTs in the CTJ region are associated with motions on Chile ridge transform faults subducted beneath South America. Continued activity of the ridge transforms beneath the fore arc enhances shear failure on adjacent portions of the interplate interface north of the triple junction. Furthermore, we show that shear stresses produced by solid Earth tides and ocean loading enhance tremor activity when they act in a plane oriented parallel to the subducted transform faults.
2 Observation and Tremor Characteristics
 In order to identify the tremor signal, we performed a day-by-day visual inspection of vertical component filtered in the 5–15 Hz range. The criteria we used to identify tremor included persistent amplitude of long-duration signals above noise, similar wave trains observed nearly simultaneously at more than one station, and correlations of visible tremor burst sequences within the longer tremor wave trains. Start and end times of the tremor signals were selected when that signal was below and above the ambient noise level. Errors in tremor onset and end times are on the order of a few seconds, well below the hourly time series analysis in this study. Cultural noise was not a concern because stations recording tremor were located in the uninhabited Chonos Archipelago.
 The CTJ region tremors are long-duration (Figure 2), semicontinuous (100–1000 min) low-frequency (1–10 Hz) seismic signals composed by several bursts (sometimes at regular time intervals; see Figure S1 in the Supporting Information1), each with a progressive amplitude sequence of increase, maximum, and decay. NVT persisted during 2 years (December 2004 to February 2007) of continuous recording on 15 stations located in the Chonos Archipelago, from a 59 station seismic network, the Chile Ridge Subduction Project (CRSP), with no more than 3 sequential days without NVT activity. Histograms of NVT occurrence at stations above the subducting Nazca plate (Figure 3) show that stations IBJ and AGU, situated above the subducted Guamblin and Darwin fracture zones, recorded minor NVT activity with duration longer than 200 min 5 times. To the south, at stations overlying the subducted Darwin and Taitao FZs, stations HUM, ISM, PLM, RPR, and NWM recorded major activity with durations longer than 1020 min 6 times and a maximum duration of 2880 continuous minutes. However, south of Taitao FZ, where the Antarctic plate subducts, NVT activity at stations HOP and SAD was rare. Particle motions of tremor bursts show that the NVTs are predominantly shear waves (Figure S2). We also observed weak, low-frequency (1–7 Hz) impulsive waves (Figure 4) within tremor signals which show typical separation times (Figure S3) and particle motions (Figure S2) of P and S waves, similar to low-frequency earthquakes (LFEs) associated with NVT in Japan [Shelly et al., 2006]. Tremor bursts are depleted in high frequencies compared to shear waves of colocated earthquakes (Figure S1).
3 Tremor Locations
 We used timing of tremor bursts recorded at four to eight stations and the source-scanning algorithm (SSA) method of Kao and Shan  to locate the tremor signal sources. The study area was parameterized as a grid of 5 × 5 km cells, using an S wave velocity model for the Patagonian Andes [Robertson et al., 2003]. The SSA then yields a tremor burst hypocentral estimate by sequential comparison of the observed and calculated NVT travel times and burst amplitudes between each potential source cell and each station where NVTs were observed. We then used a two-step procedure: In the first step, the signal was visually inspected to find a clear tremor burst. A suitable portion of tremor signal was selected if it contained a relatively high SNR, duration longer than 20 s, and if it was identified on the same seismometer components (e.g., vertical) at a minimum of four stations. The amplitude of each selected seismogram was then normalized. In the second step, the time window selected was processed through the SSA to calculate the maximum brightness function [Kao and Shan, 2004], i.e.,
where N is the number of stations, tηn is the predicted travel time from potential source-cell η to station n, τ is the origin time, un is the signal amplitude at time tηn, M is the number of samples used to compute the signal amplitude, and δt is the sampling interval. Wm is a weighting factor that depends on how much the predicted signal travel time differs from the observed arrival time. The time window mδt selected was 2.5 s. By systematically searching through all n and t in the grid for the local brightness maxima, the spatial and temporal distribution of the tremor source was found (Figure S4). Due to imperfections in the assumed velocity model, the arrival time calculated may differ from the observed time. To calibrate this error, several local earthquakes (LEs) were located using the SSA and compared with the hypocentral estimates derived using HYPO71 location software [Lee and Lahr, 1972], yielding errors of about 10 km horizontally and 30 km vertically. The low-frequency earthquakes found within the tremor signal were located using HYPO71 software when impulsive P and S waves were identifiable.
 We located a total of 343 tremors and 27 LFEs. The tremors occurred in the fore arc throughout the region of the Chonos Archipelago, west of the Liquiñe-Ofqui fault zone and the active volcanic arc, and east of the Chile trench (Figure 1). They are distributed in two clusters north of the triple junction whose centers are separated by 125 km. The denser cluster of more active, long-duration tremors is centered on the Taitao Peninsula, almost directly overlying the subducted Taitao transform fault, which is the subducted plate boundary between the Nazca and Antarctic plates, taking up some 5 cm/yr of relative slip. The tremors are distributed in the depth range of 0–90 km, peaking at 30 km depth (Figure S5), and, given poor depth control, may actually lie within the subducted NZ lithosphere and at or above the interplate contact in the overriding plate. The LFEs are scarce, but the impulsive waves allow more precise source location, demonstrating that tremor bursts and low-frequency earthquakes are colocated to within errors, and simultaneous. LFEs are distributed in the shallowest 60 km peaking at 40 km depth, which is probably a better estimation of tremor depth. The locations thus demonstrate that southern Chile NVTs are not related to the activity of the volcanic arc or the Liquiñe-Ofqui transform fault, which both lie strictly to the east of the tremors.
4 Scale Invariance of Nonvolcanic Tremors
 Regression of the time series of cumulative frequency (N) versus tremor duration (T) is better correlated assuming a power law model (R = 0.99) than an exponential model (R2 = 0.96) (Figure 5). The calculated linear model using a power law results in
 Assuming T is proportional to the seismic moment M0 [Ide et al., 2007]
we can express N in terms of M0, i.e.,
with β = 1.6 and α is some proportional constant. Assuming the relationship between scalar moment and moment magnitude [Hanks and Kanamori, 1979] is
which is similar to the Gutenberg-Richter magnitude-frequency relation for earthquakes [Gutenberg and Richter, 1945], with a b value equal to 2.4. This suggests a scale-invariance pattern similar to that observed for earthquakes and implies that tremor duration depends on stress accumulation. Typical b values for earthquakes are between 0.5 and 1.5 and are attributed to stress conditions and heterogeneities [Frohlich and Davis, 1993]; however, during earthquake swarms or volcanic sequences, that number can increase to 3.6, indicating a larger proportion of smaller than larger events, attributed to high density of cracks and/or high pore pressure [Sánchez et al., 2004].
5 Tremors and Tides
 In order to study tidal modulation of NVT activity, we calculated Fourier amplitudes of tremor occurrence time series. Tremor activity detected at two stations (ISM, PLM) on the Taitao Peninsula was binarized (1 during tremor, 0 otherwise) for each hour, 2 years of data yielding 17,520 samples. Southern Chile NVTs are clearly tidally modulated: increased tremor duration/day correlates strongly with peaks of lunar and solar tides M2, N2, K2, K1, O1, P1, and Mm, with periods of 12.421 h, 12.000 h, 11.967 h, 23.934 h, 25.819 h, 24.066 h, and 27.555 days, respectively [Wahr, 1995] (Figure 6). We do not observe the Mf fortnightly lunar (period 13.661 days) or the S2 semidiurnal (12.658 h) tidal components, whose expected amplitudes (around 2–3 times that of Mm) suggest that they should be also present. The maximum amplitudes of the tidal constituents vary, but the minimum vertical surface displacement of an observed component (Mm) is small, ~21 mm (Table S1). Evolutionary spectrograms of NVT occurrence at stations (Figures 7 and S6) confirm their periodic behavior during the 2 years of data recording, although power at expected tidal periods is much less strong for stations in the north relative to stations proximal to, or on, the Taitao Peninsula. Power spectra also demonstrate that tremor duration correlates positively with the amplitude of the tidal constituent: the greater the tidal displacement (Table S1), the greater the magnitude of the observed peak in power spectrum (Figure 6). Tidal modulation of NVT has been observed in other subduction settings, although only the M2 and N2 tidal constituents were detectable [Shelly et al., 2007b; Rubinstein et al., 2008].
5.1 Tidal Stresses on Distributed Faults of the Subducted Nazca Plate
 Solid Earth tides have been associated with seismicity on shallow faults [Heaton, 1975; Klein, 1976], and earthquake swarms along the Juan de Fuca Ridge [Tolstoy et al., 2002] suggest that faulting occurs when ocean loading is at a minimum, decreasing confining pressure and reducing normal stress. Under certain friction conditions, the projection of the solid Earth tidal stresses in the fault planes of shallow thrust earthquakes correlates with tidal maxima [Beeler and Lockner, 2003; Cochran et al., 2004]. Faults and fractures in the Nazca plate lithosphere, formed at Chile ridge spreading segments but now in the trench, are clearly observed in high-resolution bathymetry data [Ranero et al., 2006], including normal faults and fractures formed parallel to the overall ridge strike during spreading, and as multiple, complex fault surfaces generally parallel to the overall fault system strike in transform faults that offset ridge segments [Fox and Gallo, 1984; Goud and Karson, 1985; Tebbens et al., 1990; Carbotte and Macdonald, 1994; Escartín et al., 2007] Thus, subducted Chile ridge normal and transforms faults almost certainly include zones of pervasive faulting and fracturing parallel to the overall plate boundary structures (ridge segments and transforms) that can serve as shear slip surfaces and conduits for the kind of high-pressure fluids that appear to play an integral role in the modulation of nonvolcanic tremor by tides.
 In order to determine if any of these Nazca plate structures were active during NVT, we determined correlation coefficients between tremor activity and the stresses produced by solid Earth tides and ocean loading. Volumetric and horizontal shear stresses were calculated at the location of the southern tremor cluster in the Chonos Archipelago (46°S, 74.5°W) for 2 years of data (2005–2006), at 1 h sampling using the software SPOTL [Agnew, 2012]. Stresses were calculated at the surface, and we assumed that they are invariant down to 30 km where the occurrence of located tremors is maximum (Figures 8 and S5). To calculate the sea level variation, we used a global model of the oceans with a grid size of 0.125° [Matsumoto et al., 2000]. Vertical and horizontal stresses due to ocean loading were calculated using results for an infinitely long strip load (equation S1). Pacific Ocean loading was calculated assuming a loading strip 100 km wide, and the N-S trending Chonos channel system was modeled as a loading strip 5 km wide (Figure 8).
 The Nazca plate lithosphere is characterized by three types of brittle structures (Figures 1 and 8): right-lateral oceanic transform faults and normal faults that formed with the lithosphere at the Chile ridge [Cande and Leslie, 1986], and thrust faults that developed during subduction. To test the tidal effects on these structures, we calculated the shear, normal, and Coulomb stresses along representative right-lateral strike-slip with a vertical plane, normal, and reverse faults for a full range of fault orientations (Figure 8). The significance of the correlation between tremors and stresses was calculated using a Pearson correlation with a 99% level of confidence, i.e., correlation ≥0.02 (Figure 9).
 The Coulomb failure function (∆CFF), which is defined by Cocco and Rice (2002), is
where Δτ is the shear stress variation along slip on the fault, Δσn is the normal stress change on the fault, ΔP (assumed here to be 0) is the pore pressure change, and μ (0.5) is the friction coefficient.
 The highest observed correlation (0.08) occurs for shear stress acting on right-lateral strike-slip faults striking N95°E, i.e., subparallel to the subducted Taitao and Darwin transform faults and to the NZ plate convergence direction (N78°E) (Figure 9). For this type of fault, linear regression shows a coefficient of determination of 0.007, indicating that ~0.7% of tremor occurrence can be explained with a confidence level greater than 99.9 % by shear stress acting on the fault. For normal stresses, the maximum correlation (0.08) occurs for normal faults oriented N158°E, i.e., subparallel to the subducted normal faults (N168°E) associated with the Chile ridge. The maximum correlation for Coulomb stresses occurs for right-lateral strike-slip faults striking N93°E, also subparallel to the subducted transform faults. The positive correlation indicates that tremors occur more frequently when shear stresses are positive in the slip direction (Figures 10 and S7).
6 Discussion and Conclusions
 Although southern Chile NVTs present many characteristic features of NVT in the Nankai and Cascadia subduction zones (long-duration signals lasting from hours to days, depleted in high frequencies relative to earthquakes, occurring in conjunction with LFEs within the tremor signal, and tidal modulated), there are also important differences: NVTs in the Chile ridge subduction region were nearly continuous, with at most 72 h of quiescence during 2 years of recording, tremors were spatially related to the subduction of the active spreading ridge structures, and they are strongly tidally modulated, correlating with tidal constituents M2, N2, K2, K1, O1, P1, and Mm (Figure 6).
 Chilean NVT and LFE lie in a band parallel to the strike of the NZ slab, dispersed within an elongated area of 100 km width, lying 100 km east of the trench (Figure 1). The wide depth distribution of NVT (~0–90 km), with peak of occurrence at 30 km, and depths of LFE within 40 km of the surface (Figure S5) indicate that both phenomena occur mostly near the plate interface and within the overriding South American plate. Two clusters of NVT activity, one very active in the south and another less active in the north, coincide with the locations of Taitao and Darwin subducted oceanic transform faults, respectively, beneath the overriding plate. The subducted dextral Taitao transform fault separates the NZ and AN plates, is the locus of relatively rapid interplate slip at 5 cm/yr, and also generated the very slow Mw 6.9 aftershock of the great Chile earthquake studies by Kanamori and Stewart . This anomalous earthquake produced no body waves while generating a 1.5–2 h wave train of anomalously large-amplitude 10–25 s period Rayleigh waves.
 The cumulative frequency of Chilean NVT yields a relation similar to the Gutenberg-Richter law with a high b value of 2.4 (Figure 5), indicating a larger proportion of short-duration rather than long-duration events as compared to normal earthquake distributions. High b values have been observed during earthquake swarms [Holtkamp and Brudzinski, 2011] and in volcanic areas [Farrell et al., 2009; Sánchez et al., 2004]. Physical parameters influencing the b value include stress [Scholz, 1968], high pore pressure [Wyss, 1973], and density of cracks [Mogi, 1962]. If tremors are produced by superposition of many low-frequency earthquakes [Shelly et al., 2007b], this result indicates that Chilean tremors are related to the area of the main slip surface, as well as to the volume of fluids and density of cracks along the fault interface. A different result was found in the Tokai Region of Japan [Watanabe et al., 2007], where tremor amplitude and cumulative frequency of tremor duration are exponentially related, possibly as a result of changes in excess fluid pressure or variable stress drop.
 In the CTJ area, the correlation between tremor occurrence and soli-lunar tidal variations is maximum when the shear stress produced by the combination of semidiurnal, diurnal, and fortnightly tidal displacements unclamps or enhances shear on existing populations of faults in the NZ plate crust. From linear regression analysis, the highest significant correlation (R = 0.08) between tremor duration and tides occurs when tidal shear stress acts on the slip plane of a dextral strike-slip fault with a strike near N95°E (Figure 11), indicating that at a minimum ~1% of the tremors can be explained by tidally generated dextral shear stress along the Taitao and Darwin transform faults in the Nazca plate, which are loci of both low-magnitude earthquakes [Murdie et al., 1993] and anomalous slow earthquake activity in the wake of megathrusting [Kanamori and Stewart, 1979]. A similar correlation has been observed to occur for shear acting parallel to the San Andreas Fault [Thomas and Nadeau, 2009]. We also found a significant correlation (0.08), for normal stresses in a slip plane oriented N158°E with normal motion, i.e., close to the strike of normal faults formed in the subducted Nazca plate at the Chile ridge; however, for other orientations, normal faults show correlations higher than 0.05.
 The consistent relationship between the non-episodic Chilean tremors and tidal variation implies that part of NVT activity occurs when tidal stresses attain a threshold value as would be expected if the NVTs were tidally triggered by, for example, changes in Coulomb stress differences [Gomberg et al., 1998; Harris, 1998; Stein et al., 1994], resulting from the combination of ocean loading and solid Earth stresses. The apparent sensitivity of the NVT system to small stresses is consistent with tremors as a manifestation of slip on fault surfaces or fracture surfaces characterized by high density of cracks and high pore fluid pressures, even at the depth of the lower limit of subduction zone locking [Obara, 2002; Rubinstein et al., 2008; Shelly et al., 2007a; Watanabe et al., 2007].
 We also observe modulation of the NVT durations at periods of 2.6, 10.3, and 16.2 days (Figure 6), which we interpret as due to periodic infiltration and draining of the complex canal system of the archipelago associated with propagation of trapped water waves in the canals. Periodicity in channelized currents and nearshore trapped Kelvin and shelf waves is observed at the requisite frequencies in the Santa Barbara Channel (6–18 day periods) [Auad and Hendershott, 1997], in Sagami Bay (2.5–7 and >20 day periods) [Kitade et al., 1998], and offshore Honshu, Japan (3–6 day periods) [Kitade and Matsuyama, 2000]. These phenomena appear to be controlled by wind stresses in channels and across continental shelves and by variable remote adjusted sea level due to atmospheric pressure. By analogy, such periods present in NVT from Patagonian Chile are likely related to stress changes caused by periodic loading of the Chonos canals. Direct observations and/or coupled hydrodynamic-elastic modeling required to verify that such water wave phenomena occur in the complex channels of the Chonos Archipelago and could yield the observed long periods in tremor duration are beyond the scope of this paper [Zahel, 1997; Zahel et al., 2000].
 This work would not have been possible without the help of the following people: Umberto Fuenzalida, Hernan Marilao, Carmen Gloria of the Universidad de Chile; Corporación Nacional Forestal de Chile (CONAF)—Juan Fica, Claudio Manzur, Carlos Llautureo; Sra. Monica Retamal Maturana, Banco Estado; Carabineros de Chile del Region de Aysen; Armada de Chile; Cuerpo Militar de Trabajo del Ejercito de Chile (Cmdte. Roldan, Maj. Wellkner); Alcaldes de Aysen, Melinka (Luis Miranda Chiguay), Rio Ibanez, Lago Verde; Ejercito de Chile; Aeronautica de Chile (Sr. Carlos Feliu Ruiz); Automotriz Varona (Don Luis Hidalgo); Don Gustavo Lopez y Hostal Bon; Mike Fort, Noel Barstow, Bruce Beaudoin, Jim Fowler of IRIS PASSCAL; Tripulacion de LM Petrel IV (CONAF)—Juan Gallardo, Axel Hernandez, Hernan Aguilar, Jose Geicha Nauto; Gilles Rigaud, Aurelia Rigaud, Valerie Clouard, Lorena Palacio et Morgane; Eduardo Moscoso, Javier; Don Raul Hernandez, Fundo Los Nirres; Ing. Sergio Mirando Contreras (Melinka); Escuela Carlos Condell, Calete Andrade, Pto. Aguirre (Sr. Victor Figueroa); Enrique Alcalde (Cochrane); Luis Levin (Bahia Murta); Baterias GAMI (Puerto Montt); Mauricio Zambrano L. (Coyhaique Centro de Llamadas); Rolando Burgos and Roselia Delgado, Fachinal; Aladin Jara, Gob. De Chile Chico; Rolando Toloza, Min. Obras Publicas; Mark and Sra. Knipreth, Heart of the Andes Lodge; Tripulacion de El Aleph—Heraldo Zapata Rivera, Ramon Villegas, Omar Tapia Vidal, Jorge Oyarzun Inostroza; Sandalio Munoz, Fundo La Pedregosa; Don Cristian Brautigam; and the many, many people of Region XI, Aysen, who helped us enthusiastically and unstintingly and without whose help this work would have been much less fun. We are grateful to Hayward Oblad and three anonymous reviewers for useful suggestions that strengthened this paper. This work was supported by the U.S. National Science Foundation under Grant EAR-0126244 and CONICYT-CHILE under Grant 1050367.